CN113449847B - Offshore wind power rolling prediction method considering second-level time series wind speed change - Google Patents

Offshore wind power rolling prediction method considering second-level time series wind speed change Download PDF

Info

Publication number
CN113449847B
CN113449847B CN202110670786.8A CN202110670786A CN113449847B CN 113449847 B CN113449847 B CN 113449847B CN 202110670786 A CN202110670786 A CN 202110670786A CN 113449847 B CN113449847 B CN 113449847B
Authority
CN
China
Prior art keywords
wind power
wind
wind speed
data
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110670786.8A
Other languages
Chinese (zh)
Other versions
CN113449847A (en
Inventor
梅睿
顾文
袁超
唐一铭
刘亚南
杨宏宇
王斯妤
黄佳星
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Fangtian Power Technology Co Ltd
Original Assignee
Jiangsu Fangtian Power Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangsu Fangtian Power Technology Co Ltd filed Critical Jiangsu Fangtian Power Technology Co Ltd
Priority to CN202110670786.8A priority Critical patent/CN113449847B/en
Publication of CN113449847A publication Critical patent/CN113449847A/en
Application granted granted Critical
Publication of CN113449847B publication Critical patent/CN113449847B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention relates to an offshore wind power rolling prediction method considering second-level time series wind speed change, belonging to the technical field of power markets; the technical problem to be solved is as follows: the method comprises the steps of preprocessing data, processing abnormal data and completing sea wind speed and wind power normalization, further establishing a wind speed prediction model under a second-level time scale through a difference smooth power sequence, and finally establishing a rolling LSTM memory network to realize prediction of second-level time sequence data and correspond to wind power under the wind speed; the technical scheme is as follows: the offshore wind power rolling prediction method considering the second-level time series wind speed change comprises the following steps of: step S1) preprocessing wind speed and wind power data, step S2) establishing a wind speed and wind power prediction model, and step S3) analyzing and verifying an example.

Description

Offshore wind power rolling prediction method considering second-level time series wind speed change
Technical Field
The invention relates to an offshore wind power rolling prediction method considering second-level time series wind speed change, belongs to the technical field of power markets, and particularly relates to an offshore wind power rolling prediction method considering second-level time series wind speed change.
Background
The distribution is along with the continuous increase of the installed quantity of the offshore wind power, the continuous improvement of the capacity, the rapid development of the wind power generation technology and the manufacturing capability of corresponding equipment, the wind power generation becomes the most mature renewable energy with the development prospect in the prior art, the offshore wind speed is high, the single machine capacity of the fan is large, the annual operation hours can reach more than 4000 hours, the offshore wind power efficiency is 20-40% more than the annual generated energy of the onshore wind power, and the energy benefit is higher; the offshore wind power plant is far away from the land, is not influenced by city planning, does not need to worry about the influence of noise, electromagnetic waves and the like on residents, can drive the economic development of coastal areas, and is convenient for coastal heavy-load cities to consume on the spot. However, offshore wind power has the characteristics of randomness, uncontrollable property and the like, and a wind power generation company cannot report generated energy correctly, so that the reported electric quantity is too small, wind is abandoned actively, and the reported electric quantity is too large and fine, so that in practical application, wind speed needs to be predicted and short-term wind power needs to be predicted.
At present, the Prediction of offshore wind power is based on historical output data, Numerical Weather Prediction (NWP) and measured meteorological data, and a Prediction model is established to predict the future offshore wind power output. In the existing research, a new method for researching the utilization problem of the wind power fluctuation rule in ultra-short term prediction by taking the inherent fluctuation rule of the wind power as an entry point and by using a new method based on an extreme learning machine and a prediction interval formula for bootstrap is adopted to carry out wind power prediction in different seasons and verify the validity of the wind power prediction, and a wind power prediction model for supporting vector machine regression is used to effectively verify that the reliability of the wind power prediction model is respectively established on the basis of the selection of adjacent days aiming at different wind power weather types. How to enable an offshore wind power generator to plan and schedule an offshore wind power generation set by using a wind power prediction and prediction method and maximize profits of the offshore wind power generator and an electric power department urgently needs accurate wind power prediction.
However, the existing multi-factor offshore wind power prediction method cannot meet the condition of information loss, and under the condition that only the wind speed and the wind power of an offshore wind power generation set exist, how to meet the wind power prediction within several hours in the future is a research direction which is needed urgently.
Disclosure of Invention
The invention overcomes the defects of the prior art, and solves the technical problems that: the offshore wind power rolling prediction method considering the second-level time series wind speed change is characterized in that data preprocessing is utilized, abnormal data are processed, offshore wind speed and wind power normalization is completed, a wind speed prediction model under the second-level time scale is established through a difference smooth power sequence, and finally a rolling LSTM memory network is established, so that the second-level time series data are predicted and correspond to the wind power under the wind speed.
In order to solve the technical problems, the invention adopts the technical scheme that: the offshore wind power rolling prediction method considering the second-level time series wind speed change comprises the following steps of:
step S1) preprocessing wind speed and wind power data;
step S2), establishing a wind speed and wind power prediction model;
step S3) example analysis verification.
In the step S1), the preprocessing process of the wind speed and wind power data includes:
step S11) abnormal data processing and normalization:
(1) experimental data
The offshore wind farm acquires wind measurement data through various sensors, realizes rapid conversion and transmission of field data through a data transmission device, and analyzes, checks and corrects the original wind measurement data to obtain more accurate historical data;
(2) Max-Min normalization
Before the prediction model is trained, as the GRU neural unit in the model adopts Sigmoid and tanh functions as activation functions, and in order to improve the accuracy of wind power prediction and the convergence rate of data in the training process, the Max-Min normalization method is adopted to normalize the original wind power data and convert the data into data in a [0,1] interval, and the data normalization formula is as follows:
Figure GDA0003713756580000021
in the formula, y is the normalized wind power value; x is the number of max The maximum value is the maximum value in the original wind power data; x is the number of min The minimum value in the original wind power data is obtained; x is a radical of a fluorine atom i The actual wind power value is obtained;
in general, considering the power output of a wind farm as a superposition of each wind power generator, the output power of a wind turbine can be expressed by the following equation:
Figure GDA0003713756580000022
in the formula, C p The utilization coefficient of the wind energy for the fan; ρ is the air density; r is the fan blade radius; v is the wind speed; the wind energy utilization coefficient represents the ratio of wind power to wind energy, namely the conversion efficiency of the fan to the wind energy, and according to the Betz limit, the maximum wind energy utilization coefficient of the horizontal fan is 0.593 under the condition of not considering wake flow influence;
step S12) analysis of fitting relation between wind speed and wind power
Because the wind speed changes, little wind speed is unfavorable for wind power generation with too big wind speed, and little wind speed can't drive the blade and rotate, and too big wind speed can arouse marine wind power generation unit trouble, and when the design, marine wind power generation unit need install speed limiter, guarantees that the fan can the safe operation when strong wind, and following regulation is occasionally designed to the fan: cut-in wind velocity v in Cut-out wind speed v out Rated wind speed v r Therefore, the offshore wind power equation can also be expressed as:
Figure GDA0003713756580000031
wherein f (v) is an equation of the relation between the offshore wind power and the wind speed at the wind speed between the cut-in wind speed and the rated wind speed; pr represents the offshore wind power between the cut-out wind speed and the rated wind speed;
however, in reality, the relationship between the wind speed and the wind power cannot be solved accurately, and the actual equation between the wind speed and the wind power cannot be solved, so that the actual equation between the wind speed and the wind power is fitted nonlinearly by adopting a Sigmoidal model and a boltzmann equation as follows:
Figure GDA0003713756580000032
the formula (4) is a boltzmann equation, A1, A2, x0 and B are parameters of the boltzmann equation.
In the step S2), the process of establishing the wind speed and wind power prediction model is as follows:
step S21), a wind power prediction model under a second-level time scale is established:
the time sequence has dynamic time characteristics, namely the sequence value of the current moment has correlation with the sequence values of a plurality of previous moments, the correlation increases along with the reduction of time intervals, offshore wind power has a plurality of uncertain factors such as wind direction, air pressure and temperature, the future short-term wind speed change is judged according to the offshore wind power fluctuation rule and the time sequence, the wind power is predicted according to the future short-term wind speed change, and a single wind speed change prediction model under a second-level time scale is expressed as follows:
P(t)＝f 1 (P(t-θ),P(t-2θ),…)+E(t) (5)
in the formula: theta is the time interval of data acquisition; f. of 1 A time-dependent function that is an offshore wind power sequence; e (t) is prediction error at time t;
since the complexity of the weather system and the wind power sequence on the sea have unevenness, f can be reduced by differentiating the smooth power sequence 1 Complexity, reducing the prediction error, i.e.:
ΔP(t)＝f 2 (ΔP(t-θ),ΔP(t-2θ),…)+e(t) (6)
in the formula: Δ p (t) is a variation value of offshore wind power at time t and time t- θ; f. of 2 A time-dependent function that is a differential sequence of offshore wind power; e (t) is the minimum prediction error at time t; p is the offshore wind power at the time of (t-2 theta);
step S22) building a rolling LSTM neural network model:
the recurrent neural network is one of artificial neural networks, is good at processing time sequence data and can describe the data context on a time axis, the LSTM is provided by taking Hochreiter and Schmidhuber as the derivation of the recurrent neural network, a plurality of special computing nodes are added in a hidden layer of the recurrent neural network, the gradient transfer mode during back propagation is improved, the situations of gradient disappearance or gradient explosion are effectively relieved, and the problem that a prediction model of a time span cannot be established due to the long-term dependence of RNN is solved;
the influence of the time information on the previous information is controlled by introducing a gate control unit into an LSTM model network topological structure, the model has time memory and is suitable for a long-time nonlinear sequence prediction problem, the LSTM network structure is composed of an input gate, an output gate and a forgetting gate, and compared with RNN, the LSTM model topological structure is different in that: a plurality of hidden layers are arranged in the memory cell, and neurons of the hidden layers are replaced by memory cells with gating mechanisms;
obtaining a structure diagram of the LSTM memory network, wherein the memory cell unit is a core component of the LSTM memory network, and the input of the model comprises a t-time sequence input x t T-1 time hidden layer cell state h t-1 And a memory cell c t-1 (ii) a The output includes the memory cell state c t And hidden layer state h t Wherein c is t And h t The model comprises long-term and short-term memory information of the model respectively, the reading and modification of the memory cell unit are realized by controlling an input gate, a forgetting gate and an output gate, information flow between networks is carried out, tanh represents an activation function of tanh, the input gate records parameters by using the sigmoid activation function, and variables are controlled to be [0,1]]In between, realize x t To c t Control of (2); the forgetting gate selectively forgets the state of the neuron at the previous moment, and the specific expression form is to utilize a memory unit c t-1 To c t Control of (2); the output gates serving to output and control the parameter variable, i.e. using c t To h is paired with t The calculation formula is as follows:
i t ＝σ(W ix x t +W ih h t-1 +b i ) (7)
f t ＝σ(W fx x t +W fh h t-1 +b f ) (8)
o t ＝σ(W ox x t +W oh h t-1 +b o ) (9)
in the formula: i.e. i t 、f t 、o t Respectively representing the state calculation results of the input gate, the forgetting gate and the output gate; w is a group of ih 、W fh 、W ox And b i 、b f 、b o Respectively representing the weight matrix and the bias term of the corresponding gate; sigma represents a sigmoid activation function; w is a group of ih A weight matrix representing the input gate; w fh A weight matrix representing a forgetting gate; w is a group of ox A weight matrix representing the output gates;
the output result of the memory module at the time t in the LSTM model is determined by the output gate and the unit state together, and the specific formula is as follows:
c′ t ＝tanh(W c ·h t-1 +W f ·x t +b c ) (10)
c t ＝f t ⊙c t-1 +i t ⊙c t ′ (11)
h t ＝o t ⊙tanh(c t ) (12)
in the formula: c. C t ' indicating time of day t The cell state input of (1); tan h is a hyperbolic tangent activation function; w c 、b c Respectively representing the state weight matrix and the bias item of the input layer; w f A state weight matrix representing a forgetting layer; an indication that elements multiply by position;
forecasting time sequence data by rolling the LSTM memory network;
step S22) verifying the predicted performance of the rolling LSTM neural network model:
to accurately verify the predicted performance of the LSTM network model presented herein, the mean absolute percentage error y is chosen MAPE Root mean square error y RMSE And prediction accuracy y FA Model prediction as evaluation indexThe effect was analyzed, where y MAPE And y RMSE The smaller the numerical value is, the greater the goodness of fit is, the more accurate the model prediction result is, and the specific definition formula is as follows:
Figure GDA0003713756580000051
Figure GDA0003713756580000052
Figure GDA0003713756580000053
in the formula: n represents the sample capacity of the test set; x act (i) And X pred (i) And i is 1,2 and … n are the real value and the predicted value of the offshore wind power at the ith moment respectively.
In step S3), the example analysis and verification is performed by performing the derivation of the actual data in step S1) and step S2), and the result is:
y of the LSTM predictive model compared to the RNN and ARIMA predictive models MAPE Minimum, simultaneous y RMSE Lowest index, y FA The method is respectively the highest, and shows that the LSTM prediction model has a better prediction effect on the offshore wind power prediction problem of the wind speed change of the second-level time series.
Compared with the prior art, the invention has the following beneficial effects:
(1) an LSTM rolling prediction model is adopted to analyze the second-level offshore wind speed and the wind power, and the prediction of each second offshore wind power within 4 hours in the future is completed;
(2) by utilizing the characteristic that the LSTM network is suitable for the time sequence, compared with RNN and ARIMA prediction models, the prediction accuracy of the LSTM rolling prediction model constructed by the method is greatly improved;
(3) the rapid development of the computer technology is combined with the comprehensive application of a big data platform, the model is applied to other prediction fields, more effective information can be excavated, the prediction precision is further improved, and theoretical guidance can be provided for the follow-up long-term accurate prediction of the offshore wind power.
Drawings
The present invention will be described in further detail with reference to the accompanying drawings;
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is an ideal relationship between wind speed and wind power in the present invention;
FIG. 3 is a recurrent neural network architecture of the present invention;
FIG. 4 is a LSTM memory network architecture in accordance with the present invention;
FIG. 5 is a rolling LSTM memory network in accordance with the present invention;
FIG. 6 illustrates the second order wind speed and offshore wind power in an exemplary analytical embodiment of the present invention;
FIG. 7 illustrates the handling and fitting of the anomaly of the wind speed in the second class and the offshore wind power in an exemplary analysis embodiment of the present invention;
FIG. 8 is a graph of actual power curves and other model predicted power curves in an exemplary analytical embodiment of the present invention;
FIG. 9 is a graph of the actual power curve and other model predicted power curves at some time in an exemplary analytical embodiment of the present invention;
FIG. 10 shows the predicted point sample second-level power error at a portion of time in an exemplary analytical embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention; all other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention relates to an offshore wind power rolling prediction method considering second-level time series wind speed change, which comprises the following steps of:
step S1) preprocessing wind speed and wind power data;
step S2), establishing a wind speed and wind power prediction model;
step S3) example analysis verification.
In the step S1), the preprocessing process of the wind speed and wind power data includes:
step S11) abnormal data processing and normalization:
(1) experimental data
The offshore wind farm acquires wind measurement data through various sensors, realizes rapid conversion and transmission of field data through a data transmission device, and analyzes, checks and corrects the original wind measurement data to obtain more accurate historical data;
(2) Max-Min normalization
Before the prediction model is trained, as the GRU neural unit in the model adopts Sigmoid and tanh functions as activation functions, and in order to improve the accuracy of wind power prediction and the convergence rate of data in the training process, a Max-Min normalization method is adopted to normalize the original wind power data and convert the data into data in a [0,1] interval, wherein the data normalization formula is as follows:
Figure GDA0003713756580000071
in the formula, y is the normalized wind power value; x is a radical of a fluorine atom max The maximum value in the original wind power data; x is the number of min The minimum value in the original wind power data is obtained; x is the number of i The actual wind power value is obtained;
in general, considering the power output of a wind farm as the superposition of each wind turbine, the output power of a wind turbine can be expressed by the following formula:
Figure GDA0003713756580000072
in the formula, C p The utilization coefficient of the wind energy for the fan; ρ is the air density; r is windThe radius of the machine blade; v is the wind speed; the wind energy utilization coefficient represents the ratio of wind power to wind energy, namely the conversion efficiency of the fan to the wind energy, and according to the Betz limit, the maximum wind energy utilization coefficient of the horizontal fan is 0.593 under the condition of not considering wake flow influence;
step S12) analysis of fitting relation between wind speed and wind power
Because the wind speed changes, little wind speed is unfavorable for wind power generation with too big wind speed, and little wind speed can't drive the blade and rotate, and too big wind speed can arouse marine wind power generation unit trouble, and when the design, marine wind power generation unit need install the speed limiter, guarantees that the fan can the safe operation when strong wind, and following regulation occasionally is designed to the fan: cut-in wind velocity v in Cut-out wind speed v out Rated wind speed v r Therefore, the offshore wind power formula can also be expressed as:
Figure GDA0003713756580000081
wherein f (v) is an equation relating offshore wind power to wind speed at a wind speed between the cut-in wind speed and the rated wind speed; pr represents the offshore wind power between the cut-out wind speed and the rated wind speed;
however, in reality, the relationship between the wind speed and the wind power cannot be solved accurately, and the actual equation between the wind speed and the wind power cannot be solved, so that the actual equation between the wind speed and the wind power is fitted nonlinearly by adopting a Sigmoidal model and a boltzmann equation as follows:
Figure GDA0003713756580000082
the formula (4) is a boltzmann equation, A1, A2, x0 and B are parameters of the boltzmann equation.
In the step S2), the process of establishing the wind speed and wind power prediction model is as follows:
step S21), establishing a wind power prediction model under a second-level time scale:
the time sequence has a dynamic time characteristic, namely the sequence value of the current moment has correlation with the sequence values of a plurality of previous moments, the correlation increases along with the reduction of time intervals, offshore wind power has a plurality of uncertain factors such as wind direction, air pressure and temperature, future short-term wind speed change is judged according to the fluctuation rule and the time sequence of the offshore wind power, the wind power is predicted according to the future short-term wind speed change, and a single wind speed change prediction model under a second-level time scale is expressed as follows:
P(t)＝f 1 (P(t-θ),P(t-2θ),…)+E(t) (5)
in the formula: theta is the time interval of data acquisition; f. of 1 A time-dependent function that is an offshore wind power sequence; e (t) is prediction error at time t;
since the complexity of the weather system and the wind power sequence on the sea have unevenness, f can be reduced by differentiating the smooth power sequence 1 Complexity, reducing prediction error, i.e.:
ΔP(t)＝f 2 (ΔP(t-θ),ΔP(t-2θ),…)+e(t) (6)
in the formula: Δ p (t) is a variation value of the offshore wind power at time t and time t- θ; f. of 2 A time-dependent function that is a differential sequence of offshore wind power; e (t) is the minimum prediction error at time t; p is the offshore wind power at the time of (t-2 theta);
step S22) building a rolling LSTM neural network model:
the recurrent neural network is one of artificial neural networks, is good at processing time sequence data and can describe the data context on a time axis, the LSTM is provided by taking Hochreiter and Schmidhuber as derivation of the recurrent neural network, the LSTM adds a plurality of special computing nodes in a hidden layer of the recurrent neural network, improves a gradient transmission mode during back propagation, effectively slows down the situation of gradient disappearance or gradient explosion, and solves the problem that a prediction model of a time span cannot be established due to the long-term dependence problem of RNN;
the influence of the time information on the previous information is controlled by introducing a gate control unit into an LSTM model network topological structure, the model has time memory and is suitable for a long-time nonlinear sequence prediction problem, the LSTM network structure is composed of an input gate, an output gate and a forgetting gate, and compared with RNN, the LSTM model topological structure is different in that: a plurality of hidden layers are arranged in the memory cell, and neurons of the hidden layers are replaced by memory cells with gating mechanisms;
obtaining a structure diagram of the LSTM memory network, wherein the memory cell unit is a core component of the LSTM memory network, and the input of the model comprises a t-time sequence input x t T-1 time hidden layer cell state h t-1 And a memory cell c t-1 (ii) a The output includes the memory cell state c t And hidden layer state h t Wherein c is t And h t The model comprises long-term and short-term memory information of the model respectively, the reading and modification of the memory cell unit are realized by controlling an input gate, a forgetting gate and an output gate, information flow between networks is carried out, tanh represents an activation function of tanh, the input gate records parameters by using the sigmoid activation function, and variables are controlled to be [0,1]]In between, realize x t To c t Control of (2); the forgetting gate selectively forgets the state of the neuron at the previous moment, and the specific expression form is to utilize a memory unit c t-1 To c t Control of (2); the output gates serving to output and control the parameter variable, i.e. using c t To h is paired with t The calculation formula is as follows:
i t ＝σ(W ix x t +W ih h t-1 +b i ) (7)
f t ＝σ(W fx x t +W fh h t-1 +b f ) (8)
o t ＝σ(W ox x t +W oh h t-1 +b o ) (9)
in the formula: i.e. i t 、f t 、o t Respectively representing the state calculation results of the input gate, the forgetting gate and the output gate; w is a group of ih 、W fh 、W ox And b i 、b f 、b o Respectively representing the weight matrix and the bias term of the corresponding gate; sigma represents a sigmoid activation function; w is a group of ih A weight matrix representing the input gate; w is a group of fh A weight matrix representing a forgetting gate; w is a group of ox A weight matrix representing the output gates;
the output result of the memory module at the time t in the LSTM model is determined by the output gate and the unit state together, and the specific formula is as follows:
c′ t ＝tanh(W c ·h t-1 +W f ·x t +b c ) (10)
c t ＝f t ⊙c t-1 +i t ⊙c t ′ (11)
h t ＝o t ⊙tanh(c t ) (12)
in the formula: c. C t ' indicating the time of day t The cell state input of (1); tan h is a hyperbolic tangent activation function; w is a group of c 、b c Respectively representing the state weight matrix and the bias item of the input layer; w is a group of f A state weight matrix representing a forgetting layer; an element indicates that the element is multiplied by position;
the prediction of time series data is realized by rolling the LSTM memory network;
step S22) verifying the predicted performance of the rolling LSTM neural network model:
to accurately verify the predicted performance of the LSTM network model presented herein, the mean absolute percentage error y is chosen MAPE Root mean square error y RMSE And prediction accuracy y FA Analyzing the model prediction effect as an evaluation index, wherein y MAPE And y RMSE The smaller the numerical value is, the greater the goodness of fit is, the more accurate the model prediction result is, and the specific definition formula is as follows:
Figure GDA0003713756580000101
Figure GDA0003713756580000102
Figure GDA0003713756580000103
in the formula: n represents the sample capacity of the test set; x act (i) And X pred (i) And i is 1,2 and … n which are respectively the actual value and the predicted value of the offshore wind power at the ith moment.
In the step S3), the example analysis process in the embodiment is as follows:
(1) analyzing the experimental data set and the experimental environment:
the embodiment is realized in an experimental environment that an operating system is Windows 10, a memory is 8GB, a CPU is Intel CoreI i3-9100F CPU @3.60GHz and a GPU is NVIDIA GeForce GTX 1650, is developed by using Python3.8 language, experimental software platforms are Anacaoda3 and Tensorflow1.14.0, and is used for writing ARIMA commonly used for LSTM memory network, recurrent neural network and time sequence prediction.
The ideal relation curve diagram of the wind speed and the wind power in fig. 2 can also be based on abnormal data processing, in order to verify the scientificity and reliability of the offshore wind power prediction model considering the change of the wind speed of the second-order time series, the online monitoring data of the second-order wind speed and the second-order offshore wind power in one day of offshore wind power in Jiangsu province are analyzed and used by the present example, 105 offshore wind power units are totally arranged in the wind power unit group, the rated power is 1500kW, and the wind speed and the offshore wind power of one unit are shown in fig. 6.
As can be seen from fig. 6, part of the data shows that the power is less than 0, abnormal data needs to be processed, and the offshore wind power of the unit does not reach half of the rated power, so that it is determined that the relationship between the wind speed and the power is in the ascending part, and the second-level wind speed and the offshore wind power are subjected to nonlinear fitting by using a boltzmann equation, as shown in fig. 7.
And (3) adopting a difference method for wind power data, only considering the change rate in the sequence, eliminating the trend problem in the sequence, and adopting a rolling LSTM model to predict the time sequence of the offshore wind power.
The LSTM network model consists of an input layer, 1 hidden layer and an output layer, internal parameters of the LSTM are trained by adopting an Adam algorithm, an activation function in the hidden layer uses a tanh function, the rejection rate of network nodes is 0.2, in order to prevent overfitting, the iteration number is 300, the learning rate in the LSTM model is set to be 0.001, the number of neurons in the hidden layer is 4, meanwhile, the first 18 hours in one day is used as a training set, and the last 4 hours are used as a testing set.
(2) Predicting an offshore wind power result:
in the embodiment, a rolling LSTM model is selected to realize the prediction of the second-level offshore wind power, the numerical values per second of the actual power curve and the predicted power curves of other models within 4 hours are shown in FIG. 8, and the evaluation indexes of the prediction results are shown in Table 1.
TABLE 1 evaluation index of prediction results
Figure GDA0003713756580000111
By truncating the data at 22:59:30 to 23:00:30, the LSTM second prediction model is seen to be closer to the real data and the second power error is lower. The actual power curve and other model predicted power curves at the intercepted time are shown in fig. 9, and the second-level power error of the predicted point sample at the intercepted time is shown in fig. 10.
According to the above example analysis embodiment, it can be found that in step S3), the result of the example analysis verification is:
y of the LSTM predictive model compared to the RNN and ARIMA predictive models MAPE Minimum, simultaneous y RMSE Lowest index, y FA The maximum values respectively show that the LSTM prediction model has a better prediction effect on the offshore wind power prediction problem of the wind speed change of the second-level time series. Similarly, taking data of other 104 offshore wind turbine groups on the same day as an example, the LSTM prediction model is adopted to predict the offshore wind power of the wind turbine groups, and the evaluation index of the prediction result is shown in table 2.
TABLE 2 average value of evaluation indexes of prediction results of other wind turbines
Figure GDA0003713756580000121
The result shows that the prediction errors of the LSTM prediction model are lower than those of an ARIMA method commonly used for RNN and time series prediction, and the prediction stability and reliability are higher.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and these modifications or substitutions do not depart from the spirit of the corresponding technical solutions of the embodiments of the present invention.

Claims (2)

1. The offshore wind power rolling prediction method considering the wind speed change of the second-level time series is characterized by comprising the following steps of: the method comprises the following steps:
step S1) preprocessing wind speed and wind power data;
step S2), establishing a wind speed and wind power prediction model;
step S3) example analysis and verification;
in the step S1), the preprocessing process of the wind speed and wind power data includes:
step S11) abnormal data processing and normalization:
(1) experimental data
The offshore wind farm acquires wind measurement data through various sensors, realizes the rapid conversion and transmission of field data through a data transmission device, and analyzes, checks and corrects the original wind measurement data to obtain more accurate historical data;
(2) Max-Min normalization
Before the prediction model is trained, as the GRU neural unit in the model adopts Sigmoid and tanh functions as activation functions, and in order to improve the accuracy of wind power prediction and the convergence rate of data in the training process, the Max-Min normalization method is adopted to normalize the original wind power data and convert the data into data in a [0,1] interval, and the data normalization formula is as follows:
Figure FDA0003713756570000011
in the formula, y is the normalized wind power value; x is the number of max The maximum value is the maximum value in the original wind power data; x is the number of min The minimum value in the original wind power data is obtained; x is the number of i The actual wind power value is obtained;
in general, considering the power output of a wind farm as a superposition of each wind power generator, the output power of a wind turbine can be expressed by the following equation:
Figure FDA0003713756570000012
in the formula, C p The utilization coefficient of the wind energy for the fan; ρ is the air density; r is the fan blade radius; v is the wind speed; the wind energy utilization coefficient represents the ratio of wind power to wind energy, namely the conversion efficiency of the fan to the wind energy, and according to the Betz limit, the maximum wind energy utilization coefficient of the horizontal fan is 0.593 under the condition of not considering wake flow influence;
step S12) analysis of fitting relation between wind speed and wind power
Because the wind speed changes, little wind speed is unfavorable for wind power generation with too big wind speed, and little wind speed can't drive the blade and rotate, and too big wind speed can arouse marine wind power generation unit trouble, and when the design, marine wind power generation unit need install speed limiter, guarantees that the fan can the safe operation when strong wind, and following regulation is occasionally designed to the fan: cut-in wind velocity v in Cut-out wind speed v out Rated wind speed v r Therefore, the offshore wind power equation can also be expressed as:
Figure FDA0003713756570000021
wherein f (v) is an equation relating offshore wind power to wind speed at a wind speed between the cut-in wind speed and the rated wind speed; pr represents the offshore wind power between the cut-out wind speed and the rated wind speed;
however, in reality, the relation between the wind speed and the wind power cannot be solved accurately, and the actual equation between the wind speed and the wind power cannot be solved, so that the actual equation between the wind speed and the wind power is fitted nonlinearly by adopting a Sigmoidal model and adopting a boltzmann equation as follows:
Figure FDA0003713756570000022
formula (4) is a boltzmann equation; a1, A2, x 0 B is a parameter of the boltzmann equation;
in the step S2), the process of establishing the wind speed and wind power prediction model is as follows:
step S21), a wind power prediction model under a second-level time scale is established:
the time sequence has a dynamic time characteristic, namely the sequence value of the current moment has correlation with the sequence values of a plurality of previous moments, the correlation increases along with the reduction of time intervals, offshore wind power has a plurality of uncertain factors such as wind direction, air pressure and temperature, future short-term wind speed change is judged according to the fluctuation rule and the time sequence of the offshore wind power, the wind power is predicted according to the future short-term wind speed change, and a single wind speed change prediction model under a second-level time scale is expressed as follows:
P(t)＝f 1 (P(t-θ),P(t-2θ),…)+E(t) (5)
in the formula: theta is the time interval of data acquisition; f. of 1 A time-dependent function that is an offshore wind power sequence; e (t) is prediction error at time t;
since the offshore wind power sequence has unevenness due to the complexity of the weather system, f can be reduced by differentially smoothing the power sequence 1 Complexity, reducing the prediction error, i.e.:
ΔP(t)＝f 2 (ΔP(t-θ),ΔP(t-2θ),…)+e(t) (6)
in the formula: Δ p (t) is a variation value of offshore wind power at time t and time t- θ; f. of 2 A time-dependent function that is a differential sequence of offshore wind power;e (t) is the minimum prediction error at time t; p is offshore wind power at the time of (t-2 theta);
step S22) building a rolling LSTM neural network model:
the recurrent neural network is one of artificial neural networks, is good at processing time sequence data and can describe the data context on a time axis, the LSTM is provided by taking Hochreiter and Schmidhuber as derivation of the recurrent neural network, the LSTM adds a plurality of special computing nodes in a hidden layer of the recurrent neural network, improves a gradient transmission mode during back propagation, effectively slows down the situation of gradient disappearance or gradient explosion, and solves the problem that a prediction model of a time span cannot be established due to the long-term dependence problem of RNN;
the gate control unit is introduced into the LSTM network topology structure to control the influence of the current time information on the previous information, the model has time memory and is suitable for the long-time nonlinear sequence prediction problem, the LSTM network structure is composed of an input gate, an output gate and a forgetting gate, and compared with RNN, the difference is that: a plurality of hidden layers are arranged in the memory cell, and neurons of the hidden layers are replaced by memory cells with gating mechanisms;
obtaining a structure diagram of the LSTM memory network, wherein the memory cell unit is a core component of the LSTM memory network, and the input of the model comprises a t-time sequence input x t T-1 time hidden layer cell state h t-1 And a memory cell c t-1 (ii) a The output includes the memory cell state c t And hidden layer state h t Wherein c is t And h t The model comprises long-term and short-term memory information of the model respectively, the reading and modification of the memory cell unit are realized by controlling an input gate, a forgetting gate and an output gate, information flow between networks is carried out, tanh represents an activation function of tanh, the input gate records parameters by using the sigmoid activation function, and variables are controlled to be [0,1]]In between, realize x t To c t Control of (2); the forgetting gate selectively forgets the state of the neuron at the previous moment, and the specific expression form is to utilize a memory unit c t-1 To c t Control of (2); the output gates serving to output and control the parameter variable, i.e. using c t To h is paired with t The degree of influence of (a) is,the calculation formulas are respectively as follows:
i t ＝σ(W ix x t +W ih h t-1 +b i ) (7)
f t ＝σ(W fx x t +W fh h t-1 +b f ) (8)
o t ＝σ(W ox x t +W oh h t-1 +b o ) (9)
in the formula: i all right angle t 、f t 、o t Respectively representing the state calculation results of the input gate, the forgetting gate and the output gate; w is a group of ih A weight matrix representing the input gate; w fh A weight matrix representing a forgetting gate; w ox A weight matrix representing the output gates; b is a mixture of i 、b f 、b o Respectively representing respective gate bias terms; sigma represents a sigmoid activation function;
the output result of the memory module at the time t in the LSTM model is determined by the output gate and the unit state together, and the specific formula is as follows:
c′ t ＝tanh(W c ·h t-1 +W f ·x t +b c ) (10)
c t ＝f t ⊙c t-1 +i t ⊙c′ t (11)
h t ＝o t ⊙tanh(c t ) (12)
in the formula: c. C t ' indicating time of day t The cell state input of (1); tan h is a hyperbolic tangent activation function; w c 、b c Respectively representing the state weight matrix and the bias item of the input layer; w is a group of f A state weight matrix representing a forgetting layer; an indication that elements multiply by position;
the prediction of time series data is realized by rolling the LSTM memory network;
step S22) verifying the predicted performance of the rolling LSTM neural network model:
to accurately verify the predicted performance of the LSTM network model presented herein, the mean absolute percentage error y is chosen MAPE Root mean square error y RMSE And prediction accuracy y FA Analyzing the model prediction effect as an evaluation index, wherein y MAPE And y RMSE The smaller the numerical value is, the greater the goodness of fit is, the more accurate the model prediction result is, and the specific definition formula is as follows:
Figure FDA0003713756570000041
Figure FDA0003713756570000042
Figure FDA0003713756570000043
in the formula: n represents the sample capacity of the test set; x act (i) And X pred (i) And i is 1,2 and … n are the real value and the predicted value of the offshore wind power at the ith moment respectively.
2. The offshore wind power roll prediction method considering second-order time-series wind speed variations according to claim 1, characterized in that: in step S3), case analysis and verification, the actual data is nested in step S1) and step S2) to derive the result:
y of the LSTM predictive model compared to the RNN and ARIMA predictive models MAPE Minimum, while y RMSE Lowest index, y FA The maximum values respectively show that the LSTM prediction model has a better prediction effect on the offshore wind power prediction problem of the wind speed change of the second-level time series.
CN202110670786.8A 2021-06-17 2021-06-17 Offshore wind power rolling prediction method considering second-level time series wind speed change Active CN113449847B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110670786.8A CN113449847B (en) 2021-06-17 2021-06-17 Offshore wind power rolling prediction method considering second-level time series wind speed change

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110670786.8A CN113449847B (en) 2021-06-17 2021-06-17 Offshore wind power rolling prediction method considering second-level time series wind speed change

Publications (2)

Publication Number Publication Date
CN113449847A CN113449847A (en) 2021-09-28
CN113449847B true CN113449847B (en) 2022-08-19

Family

ID=77811602

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110670786.8A Active CN113449847B (en) 2021-06-17 2021-06-17 Offshore wind power rolling prediction method considering second-level time series wind speed change

Country Status (1)

Country Link
CN (1) CN113449847B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115394546B (en) * 2022-10-27 2023-01-06 山东晨宇电气股份有限公司 Offshore wind power test transformer
CN115994325B (en) * 2023-03-24 2023-05-30 湖北省气象服务中心（湖北省专业气象服务台） Fan icing power generation data enhancement method based on TimeGAN deep learning method

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108898251B (en) * 2018-06-29 2021-10-08 上海电力学院 Offshore wind farm power prediction method considering meteorological similarity and power fluctuation
CN109214566B (en) * 2018-08-30 2021-02-26 华北水利水电大学 Wind power short-term prediction method based on long and short-term memory network
CN111815033A (en) * 2020-06-19 2020-10-23 上海电力大学 Offshore wind power prediction method based on RCNN and meteorological time sequence characteristics
CN112686445A (en) * 2020-12-30 2021-04-20 南京工程学院 Photovoltaic power generation prediction method based on ARIMA-LSTM-DBN

Also Published As

Publication number Publication date
CN113449847A (en) 2021-09-28

Similar Documents

Publication Publication Date Title
Wang et al. Approaches to wind power curve modeling: A review and discussion
CN108734331B (en) Short-term photovoltaic power generation power prediction method and system based on LSTM
Zhang et al. Prediction of energy photovoltaic power generation based on artificial intelligence algorithm
Liu et al. An improved photovoltaic power forecasting model with the assistance of aerosol index data
Wang et al. Dynamic spatio-temporal correlation and hierarchical directed graph structure based ultra-short-term wind farm cluster power forecasting method
CN111753893A (en) Wind turbine generator power cluster prediction method based on clustering and deep learning
CN113449847B (en) Offshore wind power rolling prediction method considering second-level time series wind speed change
CN111695736B (en) Photovoltaic power generation short-term power prediction method based on multi-model fusion
CN113537582B (en) Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction
Goudarzi et al. Intelligent analysis of wind turbine power curve models
CN116167527B (en) Pure data-driven power system static safety operation risk online assessment method
Xu et al. Correlation based neuro-fuzzy Wiener type wind power forecasting model by using special separate signals
Huang et al. Improved clustering and deep learning based short-term wind energy forecasting in large-scale wind farms
Chen et al. Research on wind power prediction method based on convolutional neural network and genetic algorithm
CN116167531A (en) Photovoltaic power generation prediction method based on digital twin
Shabbir et al. Short-term wind energy forecasting using deep learning-based predictive analytics
Yuan et al. A PSO-LSTM model of offshore wind power forecast considering the variation of wind speed in second-level time scale
Bilal et al. Wind power conversion system model identification using adaptive neuro-fuzzy inference systems: A case study
Zhou et al. A hybrid method for ultrashort-term wind power prediction considering meteorological features and seasonal information
Yang et al. Day‐ahead wind power combination forecasting based on corrected numerical weather prediction and entropy method
Nguyen et al. A recent invasion wave of deep learning in solar power forecasting techniques using ANN
Xu et al. A classified identification deep-belief network for predicting electric-power load
Liu et al. Short-Term Wind Speed and Wind Power Prediction Based on Meteorological Model Modification
Zhu et al. Operation state of the wind turbine pitch system based on fuzzy comprehensive evaluation
Zhou et al. Day-ahead power forecasting model for a photovoltaic plant in Macao based on weather classification using SVM/PCC/LM-ANN

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant